47results about How to "Guaranteed recognition speed" patented technology

The invention discloses a facial feature recognition method and system based on the multi-region characteristic and metric learning. The method comprises the steps that convolution neural network parameters of the corresponding location and scale are obtained through the multi-scale facial area training, and corresponding facial area features are extracted according to the neural network parameters; the above features are filtered to obtain the high dimensional facial features; metric learning is conducted according to the high dimensional facial features, the after defined loss function of the feature expression is obtained through the dimension reduction processing of the features, a network model of the metric learning is obtained through the training of the loss function; the images to be recognized are inputted into the network model, the facial features are dimension reduced using the Euclidean distance in order to be recognized. In the method, multiscale is used to select multiple areas, the convolutional neural networks are trained, and the expression skills of the characteristics are improved. Meanwhile, through the selection of the acquired multi-scale features, the efficiency of expression of characteristics is improved, and the accuracy of face recognition is effectively improved.

The invention relates to the field of double dynamic license plate identification, and provides a character confidence degree-based secondary license plate identification method and apparatus for the problems existent in the prior art. The method comprises the steps of performing character identification through template matching; giving out confidence degrees of identification results; for the results with relatively low confidence degrees, performing video super-resolution processing to obtain a frame of high-quality image; and based on the image, performing secondary license plate identification through a neural network classifier. According to the method and the apparatus, a license plate confidence degree threshold Th is preset; a picture is captured from a front end and license plate locating and segmenting are performed; characters of a license plate are identified; character identification confidence degrees and license plate identification confidence degree are calculated; when the character confidence degrees are all higher than the threshold Th, a license plate identification result is directly given, otherwise, the video super-resolution processing is performed; a frame of high-quality image is obtained by utilizing time domain information; and based on the image, the to-be-identified characters are input to classifiers for performing identification according to a position relationship of the to-be-identified characters, so that a final license plate identification result is obtained.

The invention provides a mobile phone interconnection-based vehicle-mounted speech recognition system and recognition method. With the mobile phone interconnection-based vehicle-mounted speech recognition system and recognition method adopted, recognition rate and recognition speed can be ensured, and a problem of much occupation of the memory of a vehicle-mounted terminal by network resources of speech recognition can be solved. The system includes a microphone, a vehicle-mounted terminal which sends speech information to a mobile phone end and receives speech recognition result of the mobile phone end or a cloud speech recognition engine, the mobile phone end which analyzes and recognizes the speech information through mobile phone application software so as to obtain the speech recognition result of the mobile phone end, sends the speech information to the cloud speech recognition engine, receives the speech recognition result of the cloud speech recognition engine, performs arbitration on the speech recognition result of the mobile phone end and the speech recognition result of the cloud speech recognition engine through mobile phone application software, and sends the speech recognition result of the mobile phone end or the cloud speech recognition engine to the vehicle-mounted terminal according to an arbitration result, and the cloud speech recognition engine which analyzes and recognizes the speech information and sends the speech recognition result to the mobile phone end.

The invention provides a method and system for face identification. The method comprises the following steps: collecting multi-band training images of visible, near-infrared, intermediate-infrared, far-infrared, and thermal-infrared lights; carrying out evaluation and selection on extracted face image characteristics of all bands based on a sparse regularization method so as to obtain new characteristics of the face images of all the bands and corresponding new characteristic evaluation indexes after dimensionality reduction; carrying out evaluation, selection and fusion on a first characteristic set based on the parse regularization method so as to construct a second characteristic set finally expressing a face and a corresponding second characteristic evaluation index; obtaining a second characteristic set for finally expressing a face of a to-be-tested person from a third characteristic set according to a second characteristic evaluation index; and employing a nearest neighbor classifier to obtaining a classification result. According to the invention, on the basis of fusion of enough face image information, a characteristic set for finally expressing a face is ensured to have the low dimensionality, thereby ensuring the speed of face identification and the low data volume that an individual needs to store and thus improving identification precision.

The present invention relates to a key frame acquisition method for a human image video system. The method comprises the following steps of: 1) acquiring a video image, and performing key point detection on each frame of image; 2) performing attribute classification on each frame of image according to the detected key points, and calculating confidence level under each attribute; and 3) selecting P key frames according to the confidence level of each frame of image under different attributes as a basis of face recognition. Compared to the prior art, the selecting process of the key frame is improved, so that the speed is ensured and relatively high recognition performance is achieved.

The invention discloses portable deaf-mute communication equipment comprising a deaf-mute input information processing module and an able-bodied person input information processing module. A deaf-mute translates input sign language gestures or texts into voice through the deaf-mute information input processing module. An able-bodied person translates the input voice or the input texts into the sign language gestures and the texts through the able-bodied person information input processing module and displays the sign language gestures and the texts. The sign language gestures or the input texts of the deaf-mute can be converted into the voice, and the voice or the input texts of the able-bodied person can be converted into the sign language gestures and the texts. The portable deaf-mute communication equipment is equipment which can be carried by the deaf-mute and used for communicating with the able-bodied person onsite. The equipment can rapidly perform information processing and conversion and has a deep learning function and can be gradually adaptive to the expression habits of the user so that the equipment has certain real-time performance and practicality.

Embodiments of the invention provide an information processing method and apparatus as well as an information processing apparatus. The information processing method specifically comprises the steps of obtaining a food picture of a user and geographical location information corresponding to the food picture; determining a dining room corresponding to the geographical location information; performing search in dishes pictures corresponding to the dining room to obtain a result dishes picture as same as or similar to the food picture; according to a dishes name corresponding to the result dishes picture, determining a dishes name corresponding to the food picture; and outputting the dishes name corresponding to the food picture. According to the embodiments of the invention, the difficulty in information acquisition can be lowered and the efficiency of information acquisition can be improved.

The invention provides a bank card number recognition method based on OpenCV, which is used for solving the technical problem of low card number identification accuracy existing in the prior art. Therealization steps are as follows: firstly, grayscaling the bank card image, then distinguishing the strong illumination, the medium illumination, the weak illumination and the uneven illumination, andusing different binarization algorithms according to different situations to realize the pretreatment of the bank card image; the image of card number region in binary image is obtained by the methodof dilatation and erosion first and then contour extraction. Character segmentation and normalization are carried out by column projection method to realize character segmentation of binary image ofcard number area. A template matching algorithm is used to realize character recognition of normalized character image set. The invention belongs to the technical field of digital image processing, and relates to a recognition method of bank card number, which can be applied to digital recognition occasions such as license plate recognition, bill single number recognition, certificate number recognition and the like.

The invention discloses a division identification method and a device for data flow image frames, comprising the following steps: first, to preprocess the currently received data packet, and to store the corresponding image data acquired after the preprocessing of the data packet and the markers of the image data into a cache for accumulation; second, when the accumulation reaches a preset degree, to determine whether the image data accumulated in the cache forms an initial image identification region, and if so, then to identify the pattern of the image data in the initial image identification region. The invention avoids one-frame image restructuring of the received data and saves the data processing time, as well as dispenses with the pattern identification process of the needless image data, so as to ensure the identification speed; when processing high-pixel images, on the basis of ensuring the image identification quality, the invention can improve the identification rate to avoid data loss due to data accumulation.

The invention provides a non-contact palm print identity authentication method. The non-contact palm print identity authentication method comprises the following steps: acquiring a training sample image and a testing sample image, wherein the training sample image and the testing sample image comprise at least one fisted palm image and at least one unfolded palm image; reducing the resolution ratios of the training sample image and the testing sample image, and acquiring ROI of the training sample image and the testing sample image; transforming ROI of the training sample image to expand ROI of the training sample image; and extracting global features and local features which are not related to rotation, perspective and brightness transformation in ROI of the training sample image, and carrying out identity authentication according to the similarity of the global features and the local features of the testing sample image and the training sample image. Under the condition of guaranteeing recognition speed, accuracy of recognition of non-contact palm prints acquired on the basis of an intelligent mobile terminal is improved.

The invention relates to a method for automatically updating data of a training image library for face recognition. The method comprises the following steps: step 1, acquiring face picture data by using crawler technology; step 2, judging that face angle in the picture by using the three-dimensional pose angle, and filtering out the picture with too large deviation from the positive face; step 3,extracting five feature points in the picture by using the MTCNN technology, and carrying out affine transformation between the five feature points and the initially set five feature points, so as toachieve the effect of face alignment; step 4, automatically adding the face aligned picture data to the training image library. The method improves the richness and accuracy of the training data in the face recognition training process, ensures the training accuracy and training speed of the deep neural network model, and shortens the training period of the deep neural network model.

The invention discloses a fast identifying method of tag type in a radio frequency identification system. The method comprises the following steps: A. setting all possible sets of the tag type as a universal set N which comprises N elements, setting the types of tags to be identified as a set T which comprises t tags, and setting the probability of identification errors of tag types as is an element of a set; B. judging whether the radio frequency identification system is limited by length of time slot frames or not; and C. sequentially finding corresponding Xi=0 for each time slot in Z, then removing the Xi from the set N, and finally obtaining the rest element set which is the type set T of the tags to be identified, wherein the set of zero time slots is marked as Z, Xi=1 represents that i belongs to T, and Xi=0 represents that i is not belongs to T. According to the fast identifying method, the identifying speed is ensured by high probability, so that the method is a tag type identifying method suitable for large-scale radio frequency identification system, and has the characteristics of high robustness, strong expandability, small delay and the like.

The invention discloses a deep learning-based tire defect intelligent detection method. According to the method, based on an original X-ray detection device, an original video is divided into two channels through a video distributor; one channel is used for an operator to perform manual judgment, and the other channel is used for collecting an X-ray image through a high-speed video collection cardand transmitting the X-ray image to an identification system; and automation is realized by utilizing an artificial intelligence technology, so that the productivity is improved and the labor cost isreduced.

The embodiment of the invention discloses a speech recognition method and related products. The method comprises steps of acquiring speech data after a terminal device prompts to input the speech; using at least one kind of recognition algorithm to acquire at least two kinds of recognition results and outputting the at least two kinds of recognition results; and acquiring a relatively accurate recognition result in the recognition results, and determining a recognition algorithm corresponding to the relatively correct recognition result to be recognition algorithm for following speech recognition of the terminal device. According to the invention, by prompting a user to input the speech, acquiring the speech data input by the user, using one kind or multiple kinds of algorithms to obtain more than two kinds of recognition results and determining which algorithm is better according to the quite accurate recognition result confirmed by the user, the method is quite suitable for users with the private mobile phones or users with the similar accent; and under the premise that the recognition speed is ensured, the recognition accuracy of non-standard speech can be improved.

The invention relates to an infrared gesture recognition device and a recognition method. The infrared gesture recognition device comprises an infrared receiving tube, an infrared compensation tube, adriving module, a control module and a plurality of infrared emission tubes which are integrated on a PCB, wherein the infrared receiving tube, the infrared emission tubes and the infrared compensation tube are respectively arranged on the back of the PCB; open holes are formed in positions corresponding to the infrared receiving tube and the infrared emission tubes on the PCB and are used for emitting and receiving infrared ray from the front of the PCB; the infrared compensation tube is used for emitting infrared ray into the PCB. Compared with the prior art, infrared emission units and aninfrared compensation unit of the device adopt proportional circuits based on mirror current sources, so that the current in the device has certain self-regulation capability and good constant-currentcharacteristic, and the device has high temperature stability; meanwhile, compared with the prior art, the device is capable of removing the interference of the background environment light and solves the disadvantage that the infrared gesture module is influenced by the environment light at present; the reliability and the stability are improved.

The invention discloses a traffic-light driverless auxiliary device and method based on the Faster RCNN, and the device mainly comprises a main control Raspberry Pi 3B+, a wide angle camera, a buzzerprompt module, and a liquid crystal display screen module. The main control Raspberry Pi 3B+ is provided with a deep learning frame TensorFlow, and operates an RCNN (regional convolutional neural network) model. The wide angle camera collects a current road image in real time, determines the specific position of traffic lights through the deep learning model, and obtains a bounding rectangle. An ROI of the traffic lights is cut through OpenCV according to the bounding rectangle, so the calculation burden is smaller. After conventional filtering and other preprocessing operations, an image is converted into a Lab color space form a BGR color space, and channel a is independently extracted, and the traffic lights are recognized according to the ratio of the area of a connected domain. A final result will be prompted through the buzzer module and the display screen module. The device and method facilitate the reduction of traffic accidents, and improves the safety coefficient of driving.

The invention discloses a scene analysis system and method based on a deep learning technology, and method thereof. The system comprises a data acquisition subsystem and a cloud AI platform. The data acquisition subsystem acquires images and voices; a face recognition module in the cloud AI platform carries out face recognition on the to-be-detected image according to a deep learning technology; the facial expression analysis module is used for analyzing and judging expressions on the face; the voice recognition module carries out voice recognition on the to-be-detected audio according to a deep learning technology; the voice analysis module analyzes and judges semanteme and intonation of the to-be-tested audio; and the comprehensive analysis module is used for comprehensively analyzing results obtained by the facial expression analysis module and the voice analysis module. According to the method, face and voice recognition can be met at the same time, recognition results of face expressions, voice semantics and intonations are obtained according to the deep learning technology, the recognition results are more accurate, the recognition speed is guaranteed, and the scene analysis technology is further enriched.

The invention discloses an intelligent identification method for safety protection in 5T operation and maintenance, and the method comprises the following steps that a control module acquires configuration parameters and identifies a task start-stop command, wherein the configuration parameters comprise an identification time length, a camera ip address, a camera user name, a camera password, a video stream external port number, the number of pictures stored in each event and a state reporting period; one video stream management module pulls a video stream from the camera with the corresponding number according to the configuration parameters obtained from the control module; and an intelligent identification and analysis module carries out intelligent identification and event analysis onthe safety helmet and the protective clothing of the worker in the video frame. Real-time automatic identification and identification event uploading of wearing of safety helmets and protective clothing of workers on a railway 5T operation and maintenance operation site are achieved, configuration information is set remotely, identification operation tasks are started and stopped, the system operation state is monitored, and railway 5T operation and maintenance intelligent management is achieved.

The embodiments of the invention disclose a method and a device for banknote thickness abnormality identification. The method comprises the following steps: detecting the thickness of a banknote through at least two thickness sensors, wherein each thickness sensor acquires a set of first thickness information corresponding to the target points of the banknote; calculating the average value of each set of first thickness information, and getting second thickness information corresponding to each thickness sensor; and if a first proportion of the thickness sensors of which the second thickness information is within a first thickness range is less than a first proportion threshold, judging that the thickness of the banknote is abnormal. By employing the technical scheme, under the premise of ensuring the speed of banknote thickness abnormality identification, the accuracy of banknote thickness abnormality identification is increased, the time consumed by artificial identification is reduced, and the identification rate of banknotes with abnormal thickness is increased.

The invention relates to a quickly-constructed human face action unit recognition method and system. A relative distance between two corresponding key points is generated according to the reference distance between any two key points on a sample reference face, the expression distance between the two corresponding key points on an expression face and a relative relationship I between the referencedistance and the expression distance; relative angles among corresponding three key points according to the reference angles among any three key points on the reference face, the expression angles among the corresponding three key points on the expression face and a relative relationship II among the reference angles and the expression angles; therefore, the difference of AU motion amplitudes among individuals is cancelled; the difference caused by the appearance of people is balanced; on the basis of ensuring the natural appearance features of an identified person, facial action units presented in various expressions of the recognized person are recognized; subsequent AU recognition is guaranteed to be conducted accurately; construction of a feature library is proposed for the first timeaccording to the distance and angle features among face key points; the requirements for the quality and number of samples in a sample library are low; the calculation speed is high; and the parameter adjustment time is short.

The invention relates to a conveying belt tearing detection system and a conveying belt tearing detection method based on AI intelligent analysis. The conveying belt tearing detection system comprisesa light supplementing lamp, an AI sensor, an AI intelligent analysis terminal, a network switch, a background data operation terminal, a linkage output control device, alarm equipment and a shutdownprotection device. The conveying belt tearing detection method comprises the steps of: forming a belt tearing feature code through offline training and learning; deploying the belt tearing feature code to the AI intelligent analysis terminal; acquiring a video image of the AI sensor by means of the AI intelligent analysis terminal in real time; and adopting a neural network learning method to judge the length and the width of a belt crack accurately. When the belt is torn off, a shutdown protection signal and an alarm signal can be output within the shortest time, the linkage output control device is controlled according to a belt tearing detection result, the alarm signal and the shutdown protection signal are output, remote real-time protection and monitoring of the state of the conveying belt are achieved, operators in an operation room can track the production and operation dynamic state, the measurement precision is high, misoperation is avoided, and equipment linkage is facilitated.

The invention belongs to the field of disordered sorting machine vision, and particularly relates to a method for improving the recognition and positioning precision of automobile sheet metal workpieces. The method comprises the following steps of: 1, acquiring a complete scene image of a workpiece to be captured, carrying out preprocessing, completing instance segmentation on the workpiece in the image, and extracting the edge of a two-dimensional image according to an instance segmentation result; 2, preprocessing the acquired point cloud data of the to-be-captured workpiece scene, and further extracting the edge of the point cloud data by taking the edge of the extracted two-dimensional image as an index; calculating point pair features in the edge of the point cloud, and establishing global model description; and 3, carrying out online model matching, obtaining candidate poses by adopting a voting method based on a Hough voting principle, clustering the candidate poses by adopting a connectivity density clustering algorithm, and optimizing the poses by adopting an ICP registration algorithm. According to the invention, the recognition and positioning precision of the automobile sheet metal part can be greatly improved while the workpiece recognition speed is ensured.

The invention discloses a seed cotton mulching film online recognition algorithm based on hyperspectral imaging and deep learning. The algorithm includes: obtaining a reflection spectrum image of theseed cotton mulching film by using a hyperspectral imager; constructing a deep learning network consisting of a stacked weighted auto-encoder and a particle swarm optimization extreme learning machine to carry out online identification on the hyperspectral image. According to the invention, the hyperspectral images of the seed cotton mulching film are classified by using a network formed by the stacked weighted auto-encoder and the extreme learning machine in deep learning, and a weighting mechanism is introduced into each layer of auto-encoder, so that the multi-channel input advantage is ensured, and the noise influence is reduced at the same time; weights and offsets of the extreme learning machine are randomly determined, overfitting is easily generated, the weights and the offsets ofthe extreme learning machine are optimized by utilizing a particle swarm algorithm, and the classification precision is improved while the recognition speed is ensured. A deep learning network formedby the stacked weighted auto-encoder and the extreme learning machine can be used for online identification of the seed cotton mulching film.

The invention discloses a safety protection identification unit in 5T operation and maintenance. The identification unit comprises a control module, a video stream management module, an intelligent identification analysis module and a comprehensive management module. An 5T detection upper computer can set configuration information through a thrift interface of the control module and open and closean identification operation task, so the monitoring of the running state of the system is realized; the video stream management module pulls a video stream from the camera and maintains the state ofthe camera according to the configuration information and the command issued by the control module; the intelligent identification and analysis module carries out intelligent identification on operators, safety helmets and protective clothing through a safety protection identification method in 5T operation and maintenance, and carries out safety helmet and protective clothing event analysis. Andthe comprehensive management module is responsible for uploading events to the platform and periodically deleting local cache events. The safety protection identification unit realizes real-time automatic identification and identification event uploading of wearing of safety helmets and protective clothing of railway 5T operation and maintenance operation field workers.

The invention discloses a relational network-based remote sensing image small sample target identification method in a complex scenex. The invention aims to solve the problem of insufficient remote sensing ship weather scene coverage under the condition of small samples. Remote sensing ship images in five typical weather scenes are generated through style migration, an original feature and measurement module is used for carrying out fine adjustment on the remote sensing ship images in different scenes, and a fine adjustment result serves as a training set and a support set and is used for optimizing network parameters and improving the identification precision of a network. Style migration is adopted to process the remote sensing images, the weather scene coverage range of training set data and support set data is increased, high-dimensional feature vector clustering center extraction is subsequently carried out on the support set data, the calculation amount of a relational network isreduced, the accuracy of a network identification result is improved, and meanwhile, the identification speed of the network is ensured.

The invention discloses a personnel identity recognition method and system. The method comprises the steps of collecting face feature information of a recognized person; collecting finger vein featureinformation of the identified person; performing face recognition by using the face feature information to obtain a set of persons meeting a predetermined condition; and performing finger vein recognition on the personnel in the set by using the finger vein feature information to obtain a recognition result.

The invention provides a method and a system for constructing an efficient quality inspection model of a self-learning and updating training sample, wherein two SqueezeNet Pro supervised learning models are adopted, fast k. The means algorithm acts as a preselected tag; The loss function of each model is optimized by comparing the two data enhancement algorithms. Conducting self-learning and training models; And update the weights of the parent model. The invention effectively improves the identification efficiency or the identification accuracy, improves the speed of the algorithm calculation, does not increase the time complexity of the operation, and can effectively prevent the occurrence of the local optimal situation.

The invention discloses a collection system based on three-dimensional bar codes. The system comprises the following information collection processes that 1), a plurality of three-dimensional bar codes are generated, wherein the three-dimensional bar codes at least comprise reporting three-dimensional bar codes, and copy-prevention areas are set on the reporting three-dimensional bar codes; 2), users scan the reporting three-dimensional bar codes through terminals, the users verify the authenticity of the reporting three-dimensional bar codes through matching of preset rules and the copy-prevention areas, and if the reporting three-dimensional bar codes are verified successfully, the users send reporting requests; 3), after the system receives the reporting requests, the system carries outunique numbering on the users and displays preset declaration information web pages on the terminals of the users; and 4), the users fill out the declaration information web pages and submit the declaration information web pages, and the system receives form item information of the declaration information web pages filled out by the users, carries out preliminary verification on the form item information according to preset rules and uploads the form item information and unique serial numbers to a cloud platform after the preliminary verification is successful.

The invention relates to a key frame acquisition method for a portrait video system, comprising the following steps: 1) acquiring video images, and performing key point detection on each frame of images; 2) performing attribute classification on each frame of images according to the detected key points , and calculate the confidence under each attribute; 3) select the key frame of P frame according to the confidence of each frame image under different attributes, as the basis of face recognition. Compared with the prior art, the invention improves the key frame selection process, which not only ensures the speed, but also has higher recognition performance.

The invention relates to the technical field of computer vision, in particular to a seal recognition method, a device, equipment and a medium. The method comprises the following steps: collecting a to-be-recognized seal image; performing character recognition on the to-be-recognized seal image to obtain recognized characters, and performing feature vector extraction on the seal image to obtain a first feature vector; feature matching is conducted on feature vectors of existing seals and the first feature vectors, and seal characters of one or more existing seals with the matching degree exceeding a first preset threshold value are obtained; and performing similarity comparison on the seal characters of one or more existing seals and the recognition characters, outputting the seal characters with the highest similarity and exceeding a second preset threshold value as an recognition result; and when the similarity does not exceed the second preset threshold value, outputting the seal character of the existing seal with the highest matching degree as the recognition result.